1. Field of the Invention
The present invention relates to an apparatus for automatically deciding colors representing characteristics of an image.
2. Description of Prior Art
An image database system is used to store and manage a large amount of image data. In such an image database system, retrieve keys are added to each image data, and a desired image data can be retrieved by using the keys. In addition to a word, an icon or the like, a color representing characteristics of an image is used as the keys. The color means a color specified for example with hue, value and chroma (HLS space).
Previously, colors for the keys representing characteristics of images are decided manually by evaluating each image with eyes by an operator. This work for the evaluation is a large burden when a large amount of image data are dealt for an image database. Further, because the evaluation depends on judgement of the operators, the standards for deciding the colors representing characteristics of images are spread in the operators. Therefore, when such keys are used, it is afraid that proper retrieving result cannot be obtained.
In order to solve the problem, it is proposed to automatically decide colors representing characteristics of an image (hereinafter referred to as "characteristic colors"). For example, in Japanese Patent laid open Publication 5-274372/1993, all the colors in an image are classified by several representative colors, and deviations relative to the entire image are calculated for each representative color based on the frequencies thereof, and the representative color having large (or small) deviation is determined as a characteristic color of the image.
However, because all the colors in the image are classified in several representative colors and characteristic colors are selected among them, this process has a problem that if the number of representative colors is small, the number of characteristic colors to be decided is limited. On the other hand, if the process increases the number of representative colors, then the process will increase the number of characteristic colors. Since the process can utilize more characteristic colors, the process will select only colors that are similar to each other and that have large areas.